Big Data Risk Control Model Based On Federated Learning

In today's reform and opening up,small and micro enterprises have become an important force that can not be ignored in the national economy.However,due to the information asymmetry between banks and enterprises,small and micro enterprises have the problem of financing difficulties.Using big data technology to control risks has become an effective solution to solve the financing difficulties of small and micro enterprises.However,with the increasing attention to information security and privacy protection at home and abroad,the problem of data island in the financial field has become an obstacle to the development of credit risk control system.Traditional machine learning needs to gather the data together to train the model.Federated learning can use the local data distributed in each data source to update and build a shared global model by encrypting and exchanging parameters.Federated learning can not only meet the needs of multiple data sources to jointly build high-performance models,but also meet the data security and privacy protection policies.This paper uses Lending Club's loan data of small and micro enterprises to establish logistic regression models and evaluate their effects in federated learning and non-federal learning modes.The results show that the effect of Federated learning model is almost the same as that of training model with data gathered together.the performance of Federated learning model with two participants' data is better than that of local training model with only one participant's data.Therefore,federated learning,which trades less precision loss for privacy and security,can play a huge role in the field of credit risk assessment.